A multi-spectral classification scheme is proposed to identify areas with red tides through the satellite ocean color imagery by the sea-viewing wide field-of-view sensor (SeaWiFS). The eastern Seto Inland Sea, where serious red tides frequently occur, was studied. Colors of the water around a red tide or those emerging before/after that are referred to as "background ocean colors (hereafter BOCs)", and BOCs are estimated as the monthly mean of normalized water-leaving radiances (nLw) with 0.01 degree spatial resolution with SeaWiFS imagery. Criteria for detection of red tide pixels are established through analyses of the features of nLw (at 443, 490, 510, and 555nm) anomalies from BOCs and the nLw spectra together with the red tide records in Osaka Bay. The proposed scheme efficiently indicates the presence of red tides for individual match-ups with 92% accuracy, and this suggests that the scheme is appropriate to detect red tides for optically complex coastal water of Osaka Bay.